Enhancing hit identification in Mycobacterium tuberculosis drug discovery using validated dual-event Bayesian models

Area of Science:

• Computational chemistry and drug discovery
• Infectious disease research
• Machine learning applications in pharmacology

Background:

• High-throughput screening (HTS) for Mycobacterium tuberculosis (Mtb) drug discovery yields low hit rates.
• Need for improved hit identification strategies in tuberculosis (TB) drug development.
• Underutilization of computational methods in academic TB drug discovery compared to industry.

Purpose of the Study:

• To build and validate Bayesian machine learning models for predicting Mtb-active compounds.
• To enhance the efficiency of hit identification in TB drug discovery.
• To leverage publicly available HTS data for computational screening.

Main Methods:

• Developed and validated Bayesian machine learning models using large-scale HTS data.
• Computationally screened 82,403 molecules using the developed models.
• Performed in vitro assays on 550 selected molecules.

Main Results:

• Identified 124 active compounds against Mtb.
• Achieved significantly increased hit rates, ranging from 15-28%.
• Discovered FDA-approved and late-stage clinical kinase inhibitors with Mtb activity.

Conclusions:

• Bayesian machine learning models offer a highly efficient approach for Mtb drug discovery.
• The validated models and identified compounds are valuable resources for TB research.
• Computational screening significantly improves hit rates compared to traditional HTS.